Sal Khan on Civics, AI, and the Constitution
We the People
00:00
How Large Language Models Work
The way large language models work is they're trained on as much of human product knowledge as they can get access to. They also, the AI companies are paying human beings to label things and give extra context. And so they use that for things like safety and recognizing when the model might be going into not constructive places or might be getting overly biased. We've introduced notions of AI thoughts where the AI does things for itself so it can think before it speaks and then give a better response to learners.
Transcript
Play full episode
